Desktop version

Home arrow History arrow A Minimalist View on the Syntax–Semantics Relationship: Turning the Mind into a Snowflake

Invisibility issues

The issue in (7) is the visibility of the copy of XP for the labeling algorithm, a problem solved if the phase-level memory keeps the (F, F )-label visible for operations of narrow syntax. Note that the window of the phase-level memory slides down together with the inheritance of phasehood—once the phase head by inheritance takes over the function of the phase head, the structure above, labeled when the original phase head was the phase head, becomes part of the edge and is subject to inspection by the labeling procedure at the next phase level. The situation is quite different in cases in which the relevant XP occupies the edge of the original phase head, as in (8).


The edge of the original phase not being labeled at the phase level, it awaits labeling during computation of the higher phase, hence displacement of the XP simply makes it unavailable for labeling under feature sharing (with adverse consequences if the position was criterial in Rizzi’s sense, as discussed in Chomsky (2015b: 13) and Epstein, Kitahara, and Seely (2015a: 222-239)). The device of phase-level memory saves (7), to be sure, within narrow syntax; but this is not enough, which has already in itself important consequences for the syntax-semantics mapping— let it be here noted that such situation as in (7), which arises in all circumstances of A'-movement taking place subsequent to A-movement, requires not only that there be phase-level memory during derivation which allows narrow syntax to keep track of labels already assigned, so that ‘the information is available’ and ‘nothing removes it’ (Chomsky 2015b: 12); it is also necessary that the interpretive side of the derivation be given access to the information. For suppose that XP in (7) was displaced to the edge of the higher phase, beginning cyclic A-bar movement, as in (9).


During the derivation of the PH^level, the information that the copy of XP participates in (F, F )-labeling is preserved in the phase-level working memory. Once the PHj phase is completed, its complement is subject to transfer.


It should be kept in mind that the label (F, F), although still annotated explicitly in (10), is not part of the syntactic structure—it is determined by the labeling algorithm during the derivation without any tampering with the syntactic object taking place, on the most austere version of the theory without any operations on features as well, so that no ‘valuation’ of unvalued features in fact occurs, the configuration itself being sufficient for the labeling algorithm. The C-I component, on the other hand, must be able to ‘see’ the object as (F, F) as well—it is assumed that labels are required for interpretive purposes. If it does not recognize that the copy of XP is a copy, and works with the transferred structures as if it contained a head of the XP chain, on the next step, when the edge of PH0 gets transferred, it will incur a problem with the occurrence of the XP there—and the interpretation will fail. If it does see that the copy of XP is a copy, being therefore a link in a chain, it should not process it in any way other than other copies—ignoring it as possible source of a label. Phase-level memory within the derivational workspace is not in itself enough to ensure that interpretive components (the C-I component in particular, but the A-P side is equally relevant here, such labels having typically also overt effects there) are able to recover required properties of the structure. Given that (F, F )-labeling in such cases involves ‘valuation, it would be tempting to attribute the availability of A-A' chains to properties of this operation, but, as noted, if ‘valuation’ is in fact only a structural configuration in which appropriate syntactic objects appear, it is out of question to annotate it on the featural makeup of lexical heads; the same goes for any attempt to tie properties of (10) to valuation of Case, a process co-occurring with the establishment of a (ф, ф) label. One possible incarnation of such an idea would be to make the copy of XP participating in labeling under feature sharing to have the feature uCase eliminated at this step; another one would be to leave the feature uCase at this very place, stripping it off.


Although either solution makes the distinction between the two occurrences explicit (not without further problems), it is available within frameworks which allow explicit ‘tampering with features, either via heavy use of feature annotations—as the so-called minimalist grammars do (see Stabler (1997, 1998, 2001, 2011); for a treatment of such configurations see Kobele (2010a))—or at least to the extent that changes in featural compositions is allowed, so that copies of the same item may differ in their featural makeup (see Obata (2010, 2012), Obata and Epstein (2011)) for a feature-splitting approach), it is not compatible with the stance on chains which takes occurrences constituting a chain not to differ (which was a hallmark of early minimalism, where e. g. copies with unchecked formal features were to be deleted, see e. g. Nunes (2004, 2011)), except for relational properties, and there are no operations Copy or Form Chain—internal merge is all that narrow syntax has, and it does not include an operation which would ‘create copies’ of an object undergoing displacement so that they would be available for syntactic operations targeting them independently. This way of thinking about chains explains syntactic inertness of copies—they do not induce intervention effects due to their being parts of a chain, a discontinuous syntactic object, as characterized in Chomsky (2013 c, 2015b); they are invisible for the labeling algorithm and so on; but it has its consequences for cases like (10). If features are not ‘deleted’ as part of the operation of transfer, the ability of the C-I component to read off the label is not mysterious (note that assuming that the label is read off during the transfer does not change the picture: it is not syntax, but an interpretation-related mechanism that must do it, lest the hypothesis that labeling is required for interpretive purposes fall; see section 1.3.4); nor is it plausible that they become ‘invisible’ after transfer—they have to appear as properties of objects in appropriate structural configurations, so that they are interpreted as part and parcel of labeling. Their being active at a specific link of a chain requires that they be seen as belonging to one, not to an occurrence which may, but need not be, counted as a link of a nontrivial chain at the interfaces. The hypothesis that features are not deleted thus still leaves the connection between the occurrences unaccounted for; and it is not accidental that its variant emerged as an issue in accounts of phasal computation of successive cyclic movement which involved a dynamic determination of phasehood, based on the notion of convergence as relevant for the process, making it necessary to take the presence of unvalued features to suspend phasehood (see Felser (2004)): although it is couched in different terms and involves only intermediate landing sites at the edges of consecutive phases, it may be extended to cases of the kind exemplified in (10), i. e. to all intermediate chain links, and reformulated as the requirement that the interpretation of chains proceed in such a way as to preserve their being chains. In other words, just as the interpretive component should have access to properties of syntactic objects which determine how they are to be interpreted—for which aim labels are of key importance, hence features which underlie them should be accessible after the syntax-semantics transition—so should it have some (partial) access to that part of the structure which, although not yet (fully) transferred, is crucial for interpretation of the transferred chunk. In standard successive cyclic movement cases, the current analysis takes displacement to occur to avoid labeling failure in {XP, YP} configurations (more precisely, internal merge being applied freely, it takes place in such derivations which do not incur labeling failure), the copy of the displaced item being invisible for the labeling algorithm, which inspects therefore XP only and labels the structure accordingly:


When labeling in the active part of (12) takes place, the copy of YP is not visible for the labeling algorithm, with PH0 (or possibly a more complex structure, if head raising from a lower position has occurred) providing the label. Intermediate chain links in such situation seem therefore entirely irrelevant for the interpretive procedure, and this is correct as far as the labeling part is concerned—the whole point of such displacement is to ensure that the lower occurrence of YP in (12) do not participate in labeling (see also Blumel (2012)); but the lower occurrence of YP, when the whole complement of the phase is transferred to the C-I component, has equally to be seen as part of a chain. Phase-level memory solves the syntactic side of the issue; but the distinction between copies and repetitions has to be preserved after the transition to the C-I component. The lower occurrence of YP cannot be eliminated entirely, for it would severe the link between the ultimate landing site and the base position of YP, which could be separated by several phasal boundaries; it cannot be ‘annotated’ in a special way, for this would violate Inclusiveness; it cannot ‘await’ its deletion in some limbo between syntax and semantics, if computation by phases is taken seriously and the level of LF as a separate level of representation of the whole structure is eliminated (as it is officially declared to be; Martin and Uriagereka (2014) assume that whenever only parts of a chain are in a domain undergoing transfer, they are not ‘literally “handed over” to the interfaces when the domain is transferred’ (Martin and Uriagereka 2014: 176), but they do not elucidate further the mechanism of ‘not being literally handed over’)—phase-level memory has to go beyond the syntax-semantics boundary, as it were, and chains should be visible as chains for the C-I component without further ado (within structures delivered there from narrow syntax). Whereas the problem has not gone unnoticed, it is frequently seen as a ‘recombination’ problem—the issue of how the C-I component knows ‘what to join with what’, as if the transferred parts could be interpreted separately and then only combined together, with occurrences taken as relevant for semantics—basically, the ultimate landing site and the base position—taken care of. But the issue runs deeper: the C-I component must know, when the transfer of (10) occurs, that the chain does not stop within the phase (note that the presence of other features does not offer real help here: wh-features—or other features relevant for the A-bar domain—may be present on the moving syntactic object, and they may be even unvalued, as the system presented in Chomsky (2015b) requires them to be: there is still no information within the transferred chunk of the structure that displacement occurred; if it did not, and the features are unvalued, the whole structure will crash, or will ‘converge as gibberish’—but whatever happens to the structure, the C-I component must have resources to see the chain as a chain). Analogous situations will occur with short distance A- movement of IA and they crucially involve the identification of chains not merely inside narrow syntax, but at the interface and in the C-I component as well. When Ruys (2015) remarks in passing: ‘There is a technical question that all approaches discussed here (...) need to address: what triggers theX-abstraction at the landing site of movement, and how is the correct variable to abstract over selected?’ (Ruys 2015: 459 n. 4), the question touched upon is far from mere technicality, and does not depend on there being direct or indirect interpretation, or on translating movement dependencies with X-operators or otherwise: chains may cross phasal boundaries, but the C-I component has to see them as discontinuous objects, and not as separate occurrences, only later—via additional operations such as Form Chain—performing a collectio membrorum dispersa. The problem for a phasehood suspension account is that it would require that all phase boundaries be suspended whenever a displacement occurs, making the notion void of any significance; or requiring that there be massive look-ahead, again devoiding the notion of phases of real import for the computational process (see Chomsky (2000a: 107)). All this suggests that links of a chain should not be treated as purely syntax-internal formatives, with discontinuous syntactic objects constituted by them ‘collapsing’ into one or another position when it comes to the interpretation in the C-I component (or, on the picture that Martin and Uriagereka (2014) outline, developing the ideas of Uriagereka (2008), in separate C and I components, with distinct occurrences being loci of such a ‘collapse’ for different components) and eliminating them thereby from the interpretive procedure: the null minimalist hypothesis is that both features—valued and unvalued—should be visible for interpretive purposes and chain links—-bottom, intermediate and top alike—should be visible in the C-I component as well. This conforms to the status of chains as assumed in the most current minimalist setting—they are syntactic objects, albeit of a special sort: discontinuous ones, arising via multiple applications of the operation internal merge and being relevant as unified objects for the purposes of syntax, such phenomena as relativized minimality effects included: ‘The standard convention has been to take each of the copies to be an independent element. Adopting that convention, it may seem to be a stipulation to take the whole chain to be the relevant SO. But the convention has no particular merit’ (Chomsky (2013c: 44), introducing the reconceptualization of relativized minimality of Krapova and Cinque (2008); see Krapova and Cinque (2013) for some qualifications). Their being visible for the C-I component in all their glory, with intermediate links present, does not undermine the account of their origins in requirements of the computational process—derivation by phases, transfer and phase impenetrability all falling under the umbrella of the minimal search principle. That only means, however, that they do not arise in order to provide the C-I component with interpretive possibilities which chains are able to provide—not a surprising consequence in view of the non-semantics driven conception of syntax in general; what it does not mean is that they do not supply such possibilities at all. Note that this does not in turn mean that chains arise to satisfy requirements of the C-I component as a way to match syntactic structures and complexities of the semantic component: the early minimalist picture of the syntax-semantics relationship has become obsolete by now, and the working hypothesis of the present discussion is that the direction of explanatory analyses proceeds from syntactic properties and mechanisms to properties and mechanisms of the C-I component, as discussed in section 1.1.2.

< Prev   CONTENTS   Source   Next >

Related topics